Conditions At this week’s GPU technology conference, Nvidia did something we haven’t seen much from the chip designer lately: refresh a consumer product.
For the increasingly corporate-owned tech behemoth, GTC is less and less about GPUs for gamers and everything related to capitalizing on new and emerging markets like AI, robotics, autonomous vehicles, and the ever-wordy metaverse. By metaverse we mean 3D virtual reality worlds where you can interact and collaborate with simulations, applications and each other.
Nvidia CEO Jensen Huang took the stage in his signature leather jacket – or is it holodeck? we’re not sure – to reveal a trio of RTX 40-series graphics cards powered by its engineers’ Ada Lovelace architecture.
For many who followed Huang’s nearly hour and 45-minute keynote, that reveal was perhaps the only solid, relevant announcement at this fall’s event.
Using a hand-picked set of benchmarks, Huang boasted about the performance gains of the RTX 4090 and 4080 graphics cards over their predecessors. The chip designer said the RTX 4090 will deliver 2x to 4x the performance compared to the company’s previous flagship, the 3090 TI, which launched this spring.
Then there’s the price of these new RTX units. The cards are among Nvidia’s most expensive to date. At $899 for the 12GB 4080 and $1,199 for the 16GB version, the cards are $200 to $500 more expensive than the 3080 when it launched two years earlier. The price increase of the 4090 is not that big. At $1,599, it’s about $100 more than when the 3090 launched in 2020.
Huang defended the increase during a press conference on Wednesday, arguing that the performance gains and feature set more than offset the higher price. He claimed that the price increase was further justified by higher manufacturing and material costs.
“A 12-inch wafer is a lot more expensive today than it was yesterday, and it’s not a bit more expensive, it’s a ton more expensive,” he said, adding, “our performance with Ada Lovelace is monumentally better.”
But aside from the new maps, which Huang detailed in less than two minutes, business was back to normal. Here’s a recap of Nvidia’s bigger announcements at GTC.
Back to a dual architecture model
The 15 or so minutes leading up to the RTX announcement were spent on Nvidia’s new Ada Lovelace architecture, which sees the chip designer return to a dual-architecture model.
Nvidia’s previously announced Hopper architecture will power the company’s HPC and AI-focused processors like the H100, while Ada Lovelace architecture will power Nvidia’s graphics-centric chips.
Named after the 19th-century mathematician, the Ada Lovelace architecture is based on TSMC’s 4N process and features third-generation real-time ray tracing cores and fourth-generation Nv tensor cores.
So there’s the breakdown: Hooper primarily targeted high-performance computing and large AI workloads, and Lovelace primarily targeted everything else from cloud server GPUs to gaming cards.
This is hardly the first time Nvidia has used a dual-architecture model. Two generations ago, Nvidia’s data center chips, like the V100, used its Volta architecture. Meanwhile, its consumer and graphics-oriented chips, such as the RTX 2000 series and the Quadro RTX family, used the Turing microarchitecture.
In addition to Nvidia’s RTX 40 series parts, Ada Lovelace will also be powering Nvidia’s RTX 6000 series workstation cards and its L40 data center GPUs. Unlike Hopper, however, Huang says the new architecture is designed to address a new generation of graphics-centric challenges, including the rise of cloud gaming and the metaverse. These need graphics chips somewhere to render these real-time environments – with cloud gaming, the game is mainly rendered in a backend and streamed live over the internet to a screen in front of the user, e.g. B. a laptop or a phone. This saves gamers from buying and upgrading gaming rigs and/or carrying them around everywhere.
“In China, cloud gaming is going to be huge, and the reason is that there are a billion phones that game developers don’t know how to use anymore,” he said. “The best way to solve this is cloud gaming. You can reach integrated graphics and you can reach mobile devices.”
The Metaverse, but as-a-service
However, Ada Lovelace is not limited to cloud gaming applications. Nvidia is positioning the architecture as the workhorse of its first software-as-a-service offering, designed to give customers access to its Omniverse hardware and software stack from the cloud.
Omniverse Cloud provides the remote compute and software resources needed to run Metaverse applications on-demand from the cloud. The idea is that not every company wants to spend millions of dollars on one of Nvidia’s OVX SuperPods, or even has the budget to provide this level of simulation and rendering, if the Metaverse is indeed going anywhere. Instead, they can create their metaverses in Omniverse Cloud.
At the moment, Nvidia appears to be courting major logistics, manufacturing, and other industry partners, promising to help them build and visualize digital twins. These twins are large-scale simulations – each simulation is linked to the real world using real data and models – and are presented as a way to test and validate designs, processes and systems in a virtual world before introducing them to the real world World.
Yes, it’s all fancier modeling and simulation, but with new silicon, interactivity, virtual reality, and computation.
Though Omniverse Cloud is Nvidia’s first foray into managed cloud services, Huang says it won’t be the last to signal that his company is evaluating a similar model for its other software platforms.
Smarter cars, robots
Nvidia doesn’t just want to support digital twins of customers’ warehouses and production facilities. During the keynote, Huang also introduced a range of hardware designed to power everything from autonomous robots to cars.
Huang spoke about Drive Thor, Nvidia’s all-in-one computing platform designed to replace the multitude of computing systems used in vehicles today.
The technology will make its debut in China, where Nvidia says it will power the Zeekr and Xpeng 2025 vehicle line-ups, as well as QCraft’s autonomous taxi service. Of course, that’s only if US export restrictions aren’t tightened to the point where Nvidia can no longer ship — a prospect Huang downplayed during Wednesday’s press conference.
Meanwhile, Nvidia showed off its IGX and Orin Nano platforms to power the robotic minions that scurry around alongside human workers.
IGX is based on Nvidia’s previously announced Orin AGX Industrial System-on-Modules, but adds high-speed networking. According to Nvidia, one of the first applications of the board will be in surgical robots. Meanwhile, Nvidia’s Jetson Orin Nano modules are designed for less demanding applications.
Large language models for the masses
As with previous AGBs, software dominated a significant portion of the keynote. Two of the bigger releases for the fall event were Nvidia’s Large Language Model (LLM) services called NeMo and BioNeMo.
The services aim to facilitate the use of LLMs for AI researchers and biologists looking to extract insights from complex datasets. The services allow customers to integrate their existing data into customizable base models with minimal effort. For example, BioNeMo could be used to speed up protein folding research, it has been suggested.
Every single company in every single country that speaks every single language probably has dozens of different skills that their company could adapt to our large language model to run
Looking beyond the medical field, however, Huang expects LLMs to have broad applicability for the vast majority of companies. “I get the impression that every single company in every single country that speaks every single language probably has dozens of different skills that their company could adapt to our large language model in order to work,” he said.
“I’m not entirely sure how big this opportunity is, but it may be one of the greatest software opportunities of all time.”
funnel in production
Finally, Nvidia provided an update on the availability of its long-awaited Hopper H100 GPUs, which it says have entered volume production and will begin shipping to OEM system builders next month.
Announced at Nvidia’s GTC spring event, the 700W GPUs promised 6x higher AI performance compared to the outgoing A100, thanks to support for 8-bit floating point calculations. Meanwhile, Nvidia says the chip will deliver triple performance on double-precision FP64 calculations for HPC applications.
However, those hoping to get their hands on Nvidia’s in-house DGX H100 servers with their custom connectivity technology will have to wait until some point in Q1 2023, a full quarter later than expected.
While Nvidia blamed the greater complexity of the DGX system, a likely culprit is Intel’s Sapphire Rapids processors used in the systems, which were reportedly delayed until the end of the first quarter. ®
